NBER WORKING PAPER SERIES

Transcription

1 NBER WORKING PAPER SERIES CREDIT FRICTIONS AND SUDDEN STOPS IN SMALL OPEN ECONOMIES: AN EQUILIBRIUM BUSINESS CYCLE FRAMEWORK FOR EMERGING MARKETS CRISES Cristina Arellano Enrique G. Mendoza Working Paper NATIONAL BUREAU OF ECONOMIC RESEARCH 1050 Massachusetts Avenue Cambridge, MA April 2002 Comments and suggestions from Fernando Alvarez, Franz Hamann, Jonathan Heatcote, Alejandro Izquierdo, Urban Jermann, Pedro Oviedo, Fabrizio Perri, Katherine Smith, Diego Valderrama and Stan Zin are gratefully acknowledged. Mendoza also thanks the Research Department of the Inter-American Development Bank, where part of this article was written. The views expressed herein are those of the authors and not necessarily those of the National Bureau of Economic Research by Cristina Arellano and Enrique G. Mendoza. All rights reserved. Short sections of text, not to exceed two paragraphs, may be quoted without explicit permission provided that full credit, including notice, is given to the source.

2 Credit Frictions and Sudden Stops in Small Open Economies: An Equilibrium Business Cycle Framework for Emerging Markets Crises Cristina Arellano and Enrique G. Mendoza NBER Working Paper No April 2002 JEL No. F41, F32, E44, D52 ABSTRACT Financial frictions are a central element of most of the models that the literature on emerging markets crises has proposed for explaining the Sudden Stop phenomenon. To date, few studies have aimed to examine the quantitative implications of these models and to integrate them with an equilibrium business cycle framework for emerging economies. This paper surveys these studies viewing them as ability-to-pay and willingness-to-pay variations of a framework that adds occasionally binding borrowing constraints to the small open economy real-business-cycle model. A common feature of the different models is that agents factor in the risk of future Sudden Stops in their optimal plans, so that equilibrium allocations and prices are distorted even when credit constraints do not bind. Sudden Stops are a property of the unique, flexible-price competitive equilibrium of these models that occurs in a particular region of the state space in which negative shocks make borrowing constraints bind. The resulting nonlinear effects imply that solving the models requires non-linear numerical methods, which are described in the survey. The results show that the models can yield relatively infrequent Sudden Stops with large current account reversals and deep recessions nested within smoother business cycles. Still, research in this area is at an early stage and this survey aims to stimulate further work. Cristina Arellano Enrique G. Mendoza Department of Economics Department of Economics Social Sciences Building University of Maryland Duke University College Park, MD Durham, NC and NBER mendozae@econ.duke.edu

3 1 1. Introduction The severity of the financial and economic crisis that hit Mexico after the devaluation of the peso in December of 1994, and the unprecedented Tequila Effect by which Mexico s financial woes infected emerging markets worldwide were a harbinger of a period of intense turbulence in international capital markets. Seven years later, in December 2001 a major crisis broke out in Argentina with an explosive combination of sovereign default, massive currency devaluation and collapse of production and aggregate demand. In the seven years separating these two crises, similar crises engulfed nearly all of the so-called emerging markets, including Hong Kong, Korea, Indonesia, Malaysia, Thailand, Russia, Colombia, Ecuador, Brazil, and Turkey. Interestingly, devaluation itself proved not to be a prerequisite for these crises, as illustrated by the experiences of Argentina in 1995, Hong Kong in 1997 or Chile in Contagion effects similar to the Tequila Effect were also a recurrent phenomenon, as crises spread quickly to the capital markets of countries with no apparent economic linkages to countries in crisis. A favorite example is the correction in U.S. equity prices in the Fall of 1998 that resulted from the wave of margin calls on global investors triggered by the Russian default. The systemic nature of this correction forced the U.S. Federal Reserve to lower interest rates and coordinate the orderly collapse of hedge fund Long Term Capital Management. Emerging-markets crises are characterized by a set of striking empirical regularities that Calvo (1998) labeled the Sudden Stop phenomenon. These empirical regularities include: (a) a sudden loss of access to international capital markets, (b) a large reversal of the current account deficit, (c) a collapse of domestic production and aggregate demand, and (d) sharp corrections in asset prices and in the prices of non-traded goods relative to traded goods. Figures 1-3 illustrate

4 2 some of these stylized facts for Argentina, Korea, Mexico, Russia and Turkey. 1 Figure 1 shows recent time series data for each country s current account as a share of GDP. Sudden Stops are displayed in these plots as sudden, large swings of the current account that in most cases exceeded five percentage points of GDP. Figure 2 shows data on consumption growth as an indicator of real economic activity. These plots show that Sudden Stops are associated with a collapse in the real sector of the economy. Figure 3 provides information on two key financial indicators for each country, the price of domestic equity (valued in U.S. dollars) and the spread of the yield in JP Morgan s Emerging Markets Bond Index Plus (EMBI+) for each country relative to U.S. Treasury bills. Large declines in equity prices and sudden, sharp increases in EMBI+ spreads are features of Sudden Stops, with equity prices often leading the surge of the EMBI+ spread at the monthly frequency. The Sudden Stops phenomenon is seriously at odds with the predictions of the majority of existing models of current account determination and business cycles in small open economies, both frictionless Real Business Cycle (RBC) models and models with nominal rigidities. In these models, international capital markets provide the means for small open economies to borrow in order to smooth consumption when bad states of nature materialize, and to share the risk of their idiosyncratic income fluctuations with the rest of the world. A country s sudden loss of access to international capital markets is ruled out by assumption. Moreover, while RBC models have been successful at accounting for several features of regular business cycles in small open economies (see Mendoza (1991a) and (1994) and Correia, Neves 1 Calvo and Reinhart (1999) and Milesi-Ferretti and Razin (2000) document in detail the reversals of capital inflows and the sharp contractions in economic activity associated with Sudden Stops. The document by the International Monetary Fund (1999) documents the collapses in equity prices and the increase in their volatility. Mendoza (2002) and Parsley (2000) show evidence of sharp changes in the relative price of nontradables for Hong Kong, Korea and Mexico.

5 3 and Rebelo (1995)), they cannot account for the large magnitude of the collapses in output, consumption and investment, and the large adjustment in relative prices, observed during a Sudden Stop. Similarly, general-equilibrium asset pricing models have difficulties in accounting for the observed large asset price declines and for the contagion of asset price volatility across countries. The pressing need to gain a better understanding of the Sudden Stops phenomenon has led to the development of an active research program seeking to build models that can deliver predictions consistent with this phenomenon (see, for example, the November, 1996 and June, 2000 symposia issues of the Journal of International Economics or the NBER volumes edited by Edwards (2000), Krugman (2000a) and Frankel and Edwards (2002)). The initial step of this research program was to recognize that an analytical framework that aims to explain the empirical regularities of emerging markets crises requires a reconsideration of the conventional approach to model international capital markets as a perfect mechanism for consumption smoothing, risk sharing and credit allocation (see Calvo and Mendoza (1996)). Moreover, since international capital markets across industrial countries are much less volatile than emerging markets, it was also important to identify at the outset factors that may explain why emerging capital markets are significantly more prone to fail than the capital markets of industrial countries (see Krugman (2000b) and Calvo and Mendoza (2000a) and (2000b)). Thus, the common starting point of much of the literature on emerging markets crises has been to model international capital markets as subject to a variety of financial-market imperfections and to attribute these imperfections to various forms of informational frictions that are more pervasive in emerging markets than in industrial-country capital markets. Most of the literature focuses on theoretical partial-equilibrium models that yield qualitative results consistent with some of the

6 4 features that define a Sudden Stop. Very little is known yet about whether these models provide a reasonable quantitative account of a typical emerging markets crisis. The main objective of this paper is to analyze a framework that aims to integrate some of the financial-frictions channels proposed in the recent literature on emerging markets crises with an equilibrium business cycle model for small open economies. The emphasis is on developing methods for deriving the quantitative implications of the proposed framework and on using these quantitative methods to learn about the strengths and weaknesses of financial-frictions theories of the Sudden Stop phenomenon. The paper also provides a short survey of the recent literature with the aim of highlighting the differences between the alternative financial transmission mechanisms that have been proposed. The paper is organized as follows. Section 2 surveys the recent literature on creditfrictions models of Sudden Stops. Section 3 proposes a basic macroeconomic framework for integrating financial frictions into RBC models of the small open economy. Section 4 reviews four applications of this framework. Section 5 concludes and discusses several venues for further research. 2. Varieties of Credit Frictions A large portion of the recent literature on emerging markets crises proposes financial transmission mechanisms that can be roughly divided into two categories. First, a group of studies explore financial transmission mechanisms driven by a debtor s ability to pay. In these models, debtors may be willing to repay their debts but their ability to do so is threatened by the realization of bad states of nature. Creditors aim to cover their exposure to this default risk by imposing lending conditions on borrowers (usually in the form of collateral or liquidity requirements) or by choosing to incur monitoring costs to assess a borrower s claim not to be

7 5 able to repay. The second category emphasizes a debtor s willingness to pay. In these models, debtors decide optimally to renege on their debts when the expected lifetime payoff of defaulting, net of any default penalty, exceeds the expected lifetime payoff of repaying. Most of the work published to date in both of the above categories is theoretical in nature and is based in part on the related literature on financial frictions and contract theory in Macroeconomics, International Macroeconomics and Finance produced in the last two decades. Note also that the two-category classification is a natural way or organizing the literature but it hides the fact that several of the models that have been proposed combine elements of ability and willingness to pay (costly monitoring models, for example, have elements of both ability to pay and willingness to pay since they deal with the inability of a lender to tell if a defaulting borrower is unable to pay or unwilling to pay without incurring the monitoring cost). 2.1 Ability-to-pay Models Models driven by ability to pay generally specify explicit constraints linking a country s ability to acquire debt to the dynamics of income and prices or to various forms of collateral. The motivation for these constraints comes from the lenders interest in managing default risk and from informational asymmetries between borrowers and lenders. Calvo (1998) presents a very clear characterization of an ability-to-pay framework as a mechanism for explaining Sudden Stops. He considers a perfect-foresight, three-period small open economy with traded and nontraded goods that are separable in utility. An endowment of traded goods is received only in the last period and production of nontraded goods is planned at date 0, using a linear technology in which tradables enter as an input and output is obtained a period later. Thus, at date 0 firms borrow to import tradables to use as input in nontradables production acting on a perceived relative price for nontradables sold at date 1. At date 1,

8 6 tradables and nontradables are consumed, producers debt is due and new debt is contracted to import tradables for consumption. At date 2, tradables are consumed and debt is paid. In this economy, the Pareto optimal competitive equilibrium free of credit-market imperfections obtains when the date-0 perceived nontradables relative price equals the actual date-1 equilibrium price. Firms are always able to repay their debt. The outcome differs sharply if there is a temporary shock to the country s ability to access capital markets in period 1, once production plans are implemented, and there are fixed bankruptcy costs that increase the real cost of servicing debt at date 2. In this case, the competitive equilibrium yields two mutually-consistent outcomes. On one hand, there is bankruptcy, that is firms become unable to repay their debt. On the other hand, the equilibrium relative price of nontradables is lower than the perceived date-0 price. This occurs because bankruptcy costs lower the permanent income of the economy and thus the consumption of tradables at dates 1 and 2. The decline in tradables consumption at date 1 induces a fall in the marginal rate of substitution between nontradables and tradables, and hence in the equilibrium relative price of nontradables. This price decline renders firms bankrupt because they made zero-profits production plans at the higher Pareto-optimal price perceived at date 0. The shock to credit market access and the bankruptcy costs are both central to Calvo s story. If firms cannot roll over their debt temporarily but bankruptcy does not entail costs, households could borrow to cover the firms obligations (assuming unlimited-liability firms) and the Pareto-optimal equilibrium would prevail. This model is highly stylized but its main prediction (i.e., that large, unexpected relative price changes induced by credit frictions can trigger Sudden Stops) is robust to several modifications in terms of the life horizon, the form of the utility function, and the tradables endowment process. Mendoza (2002), for example,

9 7 considers an RBC-like, two-sector model that delivers similar findings by introducing a borrowing constraint set in terms of a fraction of gross domestic product in units of tradables (this model is discussed in Section 4.1 below). Ability-to-pay models of Sudden Stops based on collateral constraints typically emphasize constraints analogous to the well-known collateral constraint studied by Kiyotaki and Moore (1997). In these models, foreign debt is constrained not to exceed the discounted liquidation value of the capital stock one period into the future. Models of this class include those of Izquierdo (2000), Edison, Luangaram and Miller (2000) and Paasche (2001). Izquierdo (2000) and Edison, Luangaram and Miller (2000) use Kiyotaki-Moore (KM) constraints to study how this credit-market friction could explain the large real effects observed during the Tequila and East Asian crises. They find that the response of credit-constrained economies to financial shocks, such as a fall of the exchange rate, can greatly amplify the real effects of these shocks and lead to systemic financial collapse. Izquierdo argues that these magnification effects lead to an asymmetric response relative to what is observed when positive shocks hit the economy, and shows evidence of asymmetric responses in a panel for Latin America. Paasche (2001) examines the extent to which a crisis in a country can spread to another seemingly unrelated country, when two small open economies that face K-M constraints export differentiated goods to a large country. An adverse, temporary terms-of-trade shock triggered by a productivity shock to one of the small economies causes large capital outflows and a rapid deterioration in the current account of the other small economy. Other studies make use of collateral constraints that differ from the K-M constraint. These include Caballero and Krishnamurthy (2001), Auernhaimer and Garcia-Saltos (2000), Schneider and Tornell (1999), Christiano, Guts and Roldos (2001) and Mendoza and Smith

10 8 (2001). A common feature of these studies is that they produce collateral-based financial transmission mechanisms that magnify the real effects of macroeconomic shocks. Caballero and Krishnamurty focus on differences across domestic industries in their ability to offer useful collateral to international lenders. Auernheimer and Garcia-Saltos link the cost of borrowing explicitly to the market value of the capital stock, as a form of implicit collateral. Schneider and Tornell study the interaction of collateral constraints that comprise land, a fraction of risky capital and a fraction of bonds, with government bailouts guarantees. Christiano et al. and Mendoza and Smith study collateral constraints that depend on the current liquidation value of assets, a form of collateral constraint often referred to as a margin requirement (see Aiyagari and Gertler (1999)). Quantitative applications of ability-to-pay models in an RBC setting are rare (some examples are presented in Section 4), but there are numerical results for some of the models cited above for perfect-foresight experiments. Paasche (2001) and Auernheimer and Garcia Saltos (2000) study simulations of their models to explore the magnitude of the real effects they can produce in response to unanticipated shocks. Christiano, Gust and Roldos (2001) and Cespedes, Chang and Velasco (2001) develop quantitative models aimed at answering policy questions. Christiano et al. study how a change in the domestic interest rate affects output when a small open economy runs into a binding collateral constraint by which debt is constrained not to exceed a fraction of the current value of domestic assets. Firms require two types of working capital: domestic currency to hire domestic inputs and foreign currency to finance imports of an intermediate input. Borrowers and lenders do not anticipate the possibility of the suddenlybinding borrowing constraint hitting the economy, so the constraint emerges as an unanticipated shock to a perfect-foresight equilibrium as in the experiments of Paasche and Auernheimer and

11 9 Garcia-Saltos. In this setting, an interest rate cut can produce a fall in the value of domestic assets via a nominal exchange-rate depreciation, which in turn reduces imports of the foreign input. If the foreign input is not very substitutable for domestically-produced inputs, a contraction in output follows. Depending on how labor enters in production and how it responds to the interest rate cut, however, the model can also predict that an expansion could follow from the interest rate cut. Cespedes et al. (2001) consider a setup of monitoring costs analogous to the Bernanke- Gertler financial accelerator model to study the relation among exchange rate regimes, balancesheet effects, nominal rigidities and macroeconomic outcomes. In their model, balance-sheet effects magnify the adverse real effects of a foreign shock that triggers a real devaluation of the currency. However, these real effects are always larger with a fixed exchange rate than with a flexible one because the former does not help the economy cope with nominal rigidities and makes no difference for the nature of the Bernanke-Gertler external financing premium. From the perspective of developing an equilibrium business cycle approach to explain Sudden Stops, the above literature on ability-to-pay models faces two challenges. One is that in the majority of the existing models borrowing constraints are always binding along an equilibrium path. This rules out equilibrium dynamics in which one can observe large reversals of the current account triggered by a switch from a state of nature in which the constraint did not bind to one in which it does (as implicit in Calvo s (1998) setup). The second shortcoming is that most models deal with perfect-foresight experiments in which the credit constraint arrives as an unexpected shock. Hence, economic agents are not given the opportunity to adapt their optimal plans to the possibility of being suddenly unable to access international capital markets. As shown later, optimal plans differ sharply even if the probability of this event is negligible.

12 Willingness-to-pay Models The literature on international debt that emphasizes willingness to pay was very active in the aftermath of the developing-country debt crisis of the 1980s, and is now going through a renaissance motivated both by the emerging markets crises and the recent developments in closed-economy finance theory. The premise in this literature is that credit markets are intrinsically fragile because, in the absence of efficient mechanisms for committing debtors to fulfill their obligations, debtors optimally choose to default whenever the lifetime payoff of doing so exceeds the payoff of continuing in a credit relationship. However, the implications of this incentive-compatibility or participation constraint vary depending on the structure of the economic environment on which it is imposed. Consider first the classic article on sovereign default by Eaton and Gersovitz (1981) in which a sovereign debtor in default faces permanent exclusion of international capital markets (i.e., consumption must be set at the same level of the economy s income endowment each period). Eaton and Gerosvitz set this debtor in a global credit market with fully-informed, riskneutral lenders that are willing to lend at a default-risk premium that equates the expected return on risky sovereign debt with that of a riskless asset (up to a maximum lending ceiling). The default risk premium is an endogenous outcome that reflects the probability that borrowers find themselves in states of nature at which it is optimal to default because the participation constraint fails. Default is observed at equilibrium when these states occur. Atkeson (1991) considers a model in which the participation constraint interacts with a moral hazard problem in a contracting environment in which repayment schedules are contingent on output realizations. In particular, lenders cannot observe whether borrowed funds are used for investment or consumption. The optimal contract features capital outflows from the borrowing

13 11 country as a solution to the moral hazard problem. The above studies assume that countries cannot enter into other financial agreements after they renege on their debts, and thus that financial autarky is a credible threat. However, a key lingering question facing this literature is whether creditors really posses the ability to penalize debtors in this way. Bulow and Rogoff (1989) show that a sufficient condition for the reputational mechanism implicit in satisfying the participation constraint to fail to enforce debt repayments is that countries in default may have access to a rich set of deposit contracts with foreign creditors. Thus, a country that has large expected repayments due can default and then save the resources at stake in deposit contracts, thereby enjoying a higher level of consumption (and utility) thereafter. Rational lenders foresee this outcome and so international borrowing cannot be sustained in equilibrium. The Bulow-Rogoff paradox has been addressed in different ways. Bulow and Rogoff (1989) provided an answer based on the notion that lenders may be able to impose direct trade sanctions on defaulting borrowers. More recently, Kletzer and Wright (2000) proposed an environment in which both lenders and borrowers lack capacity of commitment. This limits the set of deposit contracts that defaulting economies have access to if they default. The authors show how the two-side commitment problem they study can restore sustainable intertemporal exchange through reputation by constructing a renegotiation proof equilibrium where payments are state contingent and contracts are incentive-compatible for both parties. Finally, Wright (2001) looks at how a country s concern for reputation can work to enforce repayment when there are also reputational incentives on the side of lenders that lead them to tacitly collude in punishing a country in default. If syndicated lending is allowed, banks collude to punish countries in default so as to preserve their own reputation for cooperation.

14 12 Quantitative applications of willingness-to-pay models applied to the study of Sudden Stops are as rare as those of the ability-to-pay models. Most of the literature focuses on examining theoretically the strategic interaction of borrowers and lenders and the properties of the resulting incentive-compatible contract, although some quantitative applications of models in this class do exist. Perri and Kehoe (2001) study business cycle co-movements across industrial countries in a two-country model in which each country is required to satisfy its participation constraint in each state of nature, following the closed-economy analysis of Kehoe and Levine (1993) and Kocherlakota (1996). From the perspective of emerging markets crises, however, this modeling approach has difficulties in accounting for defaults because the need to satisfy the participation constraints at all times rules them out at equilibrium. 2 Also, depending on preference and technology parameters and on the stochastic structure of the shocks hitting the economy, these models can predict that the incentives for default tend to be stronger for countries that are in good states of nature or that experience less macroeconomic volatility. Hamann (2002) studies the quantitative implications of the Eaton-Gersovitz framework and he does explore its ability to account for some features of emerging markets crises, albeit in the context of a pure exchange economy. He shows how increased persistence or variance in the endowment income process can reproduce some features of a Sudden Stop in which borrowing constraints are tightened and the default risk premium is increased. To do so, he exploits the feedback between these changes in the income process, the equilibrium probability of default, and the optimizing behavior of the risk-neutral lenders assumed in the Eaton-Gersovitz model. In summary, the growing literature studying emerging markets crises from the abilityand willingness-to-pay approaches has made important contributions to our understanding of the 2 In these models there can be states of nature in which borrowers and lenders agree to a zero payment but always as part of the contract that enforces the incentive compatibility constraint.

15 13 process that drives Sudden Stops. Related studies that have examined Sudden Stops as the outcome of self-fulfilling expectations, informational cascades, or working capital distortions (see, for example, Aghion, Bacchetta, and Banerjee, (2000) Calvo (2000), Chari and Kehoe (2002), Cole and Kehoe (1996), Reif (2001), Neumeyer and Perri (2001) and Oviedo (2002)) have also made valuable contributions. Yet, little is still known about the quantitative implications of these models in an equilibrium business cycle setting, and about whether the financial transmission channels they propose offer a quantitatively plausible account of the Sudden Stop phenomenon. In the next two sections we describe an equilibrium business cycle framework with financial frictions that tries to make some progress in this regard. 3. Real Business Cycles in Small Open Economies with Financial Frictions This Section proposes a modification of the standard RBC framework for the small open economy that introduces financial frictions. The standard small-open-economy RBC model (as in Mendoza (1991a)) features a representative infinitely-lived household, a representative firm operating a neoclassical production technology subject to random productivity disturbances and facing capital-adjustment costs, and an international credit market of one-period, non-contingent bonds. In this model, markets of contingent claims are incomplete but the credit market is perfect (i.e., the small open economy can borrow or lend at the world-determined real interest rate any amount that is consistent with the household s no-ponzi-game condition). In addition to the above assumptions, small open economy RBC models typically require extra assumptions regarding the nature of intertemporal preferences or international bond markets in order to support a well-defined stochastic stationary state from which to compute business cycle moments that can be compared to moments in the data. It is well-known that deterministic small-open-economy models, with standard time-separable preferences featuring

16 14 an exogenous rate of time preference and the standard global credit market with an exogenous interest rate, yield either explosive dynamics (if the rate of time preference and the interest rate differ) or stationary equilibria that depend on initial conditions (if the rate of time preference and the interest rate are identical). Obstfeld (1981) showed how preferences with an endogenous rate of time preference could be used to obtain well-defined foreign asset accumulation dynamics (with a unique steady state independent of initial conditions) in a perfect foresight small-openeconomy model. Similarly, the RBC model in Mendoza (1991a) used preferences with endogenous discounting in order to obtain a well-defined limiting distribution of foreign assets. In stochastic environments, intertemporal preferences with an endogenous rate of time preference take the form of Epstein s (1983) Stationary Cardinal Utility (SCU) function: t 1 E0 v c u c t= 0 τ = 0 exp ( τ ) ( t ) (1) In this expression, u(.) is a standard twice-continuously-differentiable and concave utility function and v(.) is the time preference function, which is assumed to be increasing, concave and twice-continuously-differentiable. SCU also imposes restrictions linking the two functions that can be interpreted as setting an upper bound on the elasticity of the rate of time preference with respect to the argument of utility. These restrictions imply that the impatience effect, by which changes in date-t consumption alter the rate at which all future utility flows are discounted, must be small (in the sense that the rate of time preference, exp(v(.)), is increasing in the argument of utility but maintaining the condition that u 1 ( )exp( v ( )) is non-increasing). This is necessary in order to ensure that consumption at every date is a normal good and that the model supports a well-defined unique, invariant limiting distribution (see Epstein (1983)). Endogenous discounting is not the only method used to obtain a unique, invariant

17 15 limiting distribution in stochastic small-open-economy models. Other methods explored in the literature include finitely-lived households (i.e., Blanchard preferences), transactions costs in foreign assets, and ad-hoc interest rate functions that link the rate of interest to the stock of foreign debt. Moreover, since the small-open-economy RBC model has only one non-statecontingent asset and hence features non-insurable, idiosyncratic risk, the model is in the class of incomplete-markets models for which precautionary savings can yield a well-defined stochastic steady state even with an exogenous discount factor (as long as the rate of time preference is set higher than the world interest rate, see Chapter 14 in Ljungqvist and Sargent (2000)). Some of these alternative methods can yield similar results as the endogenous discount factor in particular log-linear quantitative applications (see Schmittt-Grohe and Uribe (2001)), but this equivalence is unlikely to hold in models that aim to capture a Sudden Stop s large, nonlinear adjustments. In addition, Epstein s utility function is the only formulation consistent with three standard features of RBC models: economic agents are infinitely-lived, the rate of time preference and the rate of interest are equal in the long run, and the long-run foreign asset position is determined by the interaction of household preferences with the real interest rate and not by ad-hoc specifications of transactions-costs or interest-rate functions. Moreover, Epstein (1983) showed that SCU as a representation of a preference order requires weaker axioms of consumer theory than the ones required by time-separable utility with exogenous discounting. In the context of models with financial frictions, preferences with endogenous discounting have the extra advantage that they allow for the possibility that credit constraints can remain binding at steady state. This is because a binding credit constraint drives a wedge between the intertemporal marginal rate of substitution in consumption and the rate of interest. In a stationary state with a binding credit constraint the rate of time preference adjusts

18 16 endogenously to accommodate this wedge. In contrast, in models with an exogenous discount factor credit constraints never bind in the long run (if the exogenous rate of time preference is set greater or equal than the world interest rate) or must always bind at steady state (if the rate of time preference is fixed below the interest rate). The recursive representation of the competitive equilibrium of the small-open-economy RBC model is characterized by optimal decision rules for the vector y [ c, n, i] = of control variables (where c is consumption, n is labor supply, and i=k -k(1-δ) is gross investment) and the vector x [ k, b ] = of endogenous states (where k is capital accumulation and b is foreign asset accumulation) that solve the following dynamic-programming form of a planner s problem: k, b, c, n { [ ]} V( k, b, e) = max u( c, n) + exp( v( c, n)) E V( k, b, e ) s.t. c = ef( k, n) k + k(1 δ) ϕ( k, k ) b + br (2) where f(.) is a neoclassical production function, δ is the rate of depreciation of the capital stock, ϕ is a convex capital-adjustment-cost function, R is the world s gross real interest rate and e is a productivity disturbance that follows a Markov process with a known transition probability matrix. The initial state of the system is given by the observed realization of e and by the vector [ k, b] x =, where k is the initial capital stock and b the initial net foreign asset position. The above social planner s problem can be decentralized in standard fashion to establish the equivalence between the planner s problem and the competitive equilibrium of an economy in which households own the factors of production and rent them out to profit-maximizing firms. The prices and factor rental rates that support equilibrium allocations include the wage rate, w, the rental rate of capital, r, and the relative price of investment goods in terms of consumption goods, q which is not equal to 1 because of the capital-adjustment costs. The price vector that

19 17 households and firms take as given is thus defined by p [ w, r, q] =. Consider now a modification of the above economic environment in which financial markets are imperfect. The first change to notice is that, since financial frictions distort the competitive equilibrium, it may be necessary to study the equilibria of these economies directly in decentralized form. Alternatively, it may be possible to construct a variant of a planning problem that captures the distortions induced by specific forms of financial-market imperfections on the competitive equilibrium (as in the applications studied in Section 4). This paper focuses on financial frictions that reduce to functional constraints on the small open economy s ability to contract foreign debt of the following form: b h( k, x, y, e, p) (3) This state-contingent borrowing constraint is not formally derived as a feature of an optimal credit contract between the small open economy and its international creditors. However, the next section s review of four applications of this framework relates explicit forms of the borrowing constraint to existing results for optimal credit contracts or to observed practice in actual credit markets. The above specification is general enough to include a broad range of applications, including liquidity requirements, margin constraints, borrowing ceilings set to support debt repayment incentives, and collateral constraints set to the current liquidation value of assets. One exception is the Kiyotaki-Moore collateral constraint, which depends on future realizations of an element of p (the next period s equity price). 3 The appropriate planner s-problem representation of the borrowing constraint can be derived once explicit functional forms for the financial frictions affecting households and/or 3 Despite this limitation, other collateral constraints that are included in h, like margin requirements, have similar asset-pricing implications as Kiyotaki-Moore constraints (see Mendoza and Smith (2001) for details).

20 18 firms in the decentralized competitive equilibrium are defined. For example, the planner s borrowing constraint may represent a constraint limiting households foreign debt not to exceed a fraction of factor income or firms ability to leverage their debt on the value of their capital. Under particular assumptions, the competitive equilibria of economies with these constraints can be shown to be equivalent to that of a planning problem in which debt is limited not to exceed a fraction of output or a fraction of the market value of the capital stock. In general, however, whether such equivalence between competitive equilibria and a planner s problem exists needs to be examined case by case. At this level of generality it is possible to extract some important properties of smallopen-economy RBC models that incorporate the above borrowing constraint: (1) The borrowing constraint is occasionally binding. Whether the constraint binds or not depends ultimately on the initial state (x,e). Note that since the equilibrium is represented in recursive, social-planner s form, the optimal decision rules for k and y and the equilibrium pricing vector p in the right-hand-side of (2) are also functions of x and e. (2) The dynamics of the borrowing constraint and equilibrium allocations and prices feature endogenous feedback effects. Knowledge that the constraint is a feature of the financial markets in which the small open economy participates, influences the optimal forwardlooking plans that economic agents formulate. As a result, it also influences the equilibrium dynamics of prices and allocations, which in turn determine whether the constraint binds or not. In particular, the agents expectations that they may face future states in which the borrowing constraint can bind influences their decisions to accumulate assets today and thus the probability that the constraint can actually bind in the future. (3) Suddenly-binding borrowing constraints can produce non-linear dynamics and country-

21 19 specific risk premia. When the economy switches from a state in which the constraint does not bind to a state in which it binds, there are discrete jumps upward in the effective intertemporal relative price of current consumption and downward (upward) in the level of current (future) consumption. This follows from the household s consumption Euler equation, which takes the form: [ R ] λ = E λ + µ (4) where λ is the lifetime marginal utility of consumption (i.e. the non-negative Lagrange multiplier on the resource constraint facing the planner) and µ is the non-negative Lagrange multiplier on the borrowing constraint. The increased effective real interest rate in the states in which the borrowing constraint binds represents an endogenous, country-specific risk premium on external financing. (4) The borrowing constraint can depress asset prices and trigger Fisherian debt-deflation dynamics. As shown in the next Section, the higher effective interest rates implied by either a currently-binding or expected-future-binding borrowing constraint depresses current equity prices by increasing the discount rates relevant for discounting dividend streams in the valuation of assets. If the specification of h(.) links the borrowing constraint to asset prices, the adverse asset-pricing implications of the constraint add a feedback effect in the spirit of Fisher s debt deflation mechanism: an initial shock triggers the borrowing constraint, this leads to a decline in asset prices, which then leads to an even tighter borrowing constraint and thus a downward spiral on asset prices and access to foreign financing. An important implication of these four properties is that the small-open-economy RBC framework with borrowing constraints is endowed with a self-adjustment mechanism that

22 20 actually works to weaken the quantitative effects of financial frictions. That mechanism is the precautionary savings motive: Risk-averse agents respond to their non-insurable exposure to the risk of a binding borrowing limit, which adversely affects their consumption plans, by accumulating a buffer stock of assets. Precautionary savings can also distort portfolio choice in models in which foreign assets co-exist with other vehicles of saving such as physical capital or equity. As mentioned earlier, this precautionary-savings effect is already present in the frictionless small-open-economy RBC model because the non-contingent international bond cannot fully insure agents against country-specific idiosyncratic income shocks. The effect is stronger in the presence of borrowing constraints, however, because agents in the frictionless model could rely on a perfect international credit market to support optimal consumption plans and they have the option of using domestic investment as an alternative savings vehicle. 4 Dealing with the precautionary savings motive is an important unfinished task for the literature on financial frictions in small open economies. The majority of models of the Sudden Stops phenomenon abstract from it by assuming perfect foresight or by focusing on experiments in which the financial frictions are a negative unanticipated shock to the economic environment. Thus, in these experiments economic agents are not allowed to condition their behavior on the possibility of the occurrence of states of nature in which they are forced out of international capital markets. While this assumption has proven useful for obtaining insightful analytical results, it is important to relax it in order to integrate financial frictions into equilibrium business cycle models for small open economies. In principle, precautionary savings may seem reason to cast doubt on the potential of models with financial frictions to offer a quantitatively significant account of the empirical 4 Mendoza (1991b) shows that, in the standard small open economy RBC model, agents can adjust investment and suffer a negligible welfare loss even if they are forced out of the world

23 21 regularities of emerging markets crises. Yet, precautionary savings are in fact a key property of these models that enables them to mimic an observed feature of the data: Sudden Stops are dramatic but relatively rare. The cyclical downturns experienced in countries that have suffered Sudden Stops were large, but they were also abnormally large relative to those countries regular business cycles (see Mendoza (2002) and Calvo and Reinhart (2001)). Thus, it seems natural to think of designing equilibrium business cycle models that aim to explain Sudden Stops by nesting these dramatic, rare events within more a regular business cycle pattern. 5 It is undesirable to follow an approach with two disconnected theories of economic fluctuations for emerging economies, one for Sudden Stops and one for tranquil times, in which key assumptions regarding the ability of agents to condition their behavior on the possibility of the economy suddenly losing access to world capital markets in the future are emphasized in the former but dismissed in the latter. 4. Applications This Section of the paper reviews four applications of the general framework proposed in Section 3 that aim to capture financial frictions similar to those emphasized in the literature on emerging-markets crises. The first application considers a liquidity requirement that at equilibrium represents a constraint on the debt-to-output ratio of a two-sector economy (this model is borrowed from Mendoza (2002)). The liquidity requirement incorporates the phenomenon known as liability dollarization: foreign debt is denominated in units of tradable goods but serviced in part with income generated in the nontradables sector, and hence large credit market altogether. 5 If Sudden Stops are nested as rare events within regular business cycles, the RBC approach to compare statistical moments of detrended data with the moments of the ergodic distribution of business cycle models is unlikely to be useful for studying Sudden Stops. The next Section discusses one alternative approach that can be followed to study the models ability to explain Sudden Stops.

24 22 swings in the relative price of nontradables, or in the real exchange rate, can trigger binding borrowing constraints and sudden current account reversals. The second application reviews the asset-pricing model of Mendoza and Smith (2001). This model incorporates two financial frictions. First, a friction in the international credit market analogous to a margin requirement by which a fraction of the value of equity holdings must be offered as collateral for foreign debt. Second, informational frictions in the equity market that result in transactions costs in trading the small open economy s equity for foreign securities firms. The third application looks at Arellano s (2002) analysis of borrowing constraints that enforce credit-market participation constraints in an environment with incomplete insurance markets. The fourth application considers a model with capital adjustment costs that features a collateral constraint similar to the margin requirement used in the asset-pricing model but in a setup with endogenous capital accumulation. Foreign creditors retain as collateral a fraction of the value of the economy s capital stock where this value is determined by Tobin s q. 4.1 Liquidity Requirements and Liability Dollarization in a Two-Sector Economy Consider a small open economy with two sectors, a tradable goods sector and a nontradable goods sector. The output of tradables is a stochastic endowment exp(ε)y T, where ε is random variable that follows a Markov process. Non-tradable goods are produced using a Cobb- Douglas production function 1 Fkn (, ) exp( ε) Ak α n α =, where A is a productivity scaling factor, n is variable labor input, k is the time-invariant capital stock, and 0<α<1 is the output share of labor income. The production of nontradables is subject to productivity shocks, which are assumed to be perfectly correlated with the shocks to the tradables endowment for simplicity. Private consumption of tradables (c T ) and nontradables (c N ) is aggregated into a composite good using a standard constant-elasticity-of-substitution (CES) aggregator

25 23 1 η η η T N ( ) (1 ω) ( ) c= ω c + c, where 0<ω<1 and η>-1 are the standard CES parameters (1/(1+η) is the elasticity of substitution between c T and c N ). The argument of utility adopts the Greenwood-Hercowitz-Huffman specification, δ n c (with δ> 1), which makes the marginal δ rate of substitution between n and c independent of the latter and sets the wage elasticity of labor supply equal to 1/(δ-1). The economy has access to a global credit market of one-period bonds (b) in which it acts as a small price-taker. The world gross real interest rate exp( ε R ) R is subject to random shocks R ε which also follow a Markov process. The government of the small open economy levies a consumption tax τ that is uniform across goods. This tax is intended to capture some of the distortions that can result from fiscal, monetary and exchange-rate policies in models in which money enters as an argument of the utility function or as a means to economize transactions costs (see Mendoza (2001) and Mendoza and Uribe (2001) for details). Private agents perceive tax policy as uncertain and thus attach positive probability to scenarios in which the economy can switch from low to high-tax regimes and vice versa. Hence, the consumption tax is modeled as a standard regime-switching, asymmetric Markov process. The government also maintains a constant level of unproductive purchases of nontradables financed by a time-invariant lump-sum tax N T. This ensures that the dynamics of the relative price of nontradables (p N ) are driven only by changes in private-sector supply and demand choices and not by endogenous changes in government purchases due to fluctuations in tax revenue. Fluctuations in tax revenue result in fluctuations of unproductive government purchases of tradable goods around a trend level financed also by a time-invariant